KG-DTA: A knowledge graph-based meta-path learning framework to predict drug-target binding affinity
Document Type
Conference Proceeding
Publication Date
3-5-2025
Abstract
Drug target affinity (DTA) prediction is essential at different phases of drug discovery and virtual drug screening. However, traditional wet lab experiments are time and resource-intensive owing to the complex interactions within the vast biological and chemical spaces. Computational methods have proven their potential to predict DTA, but none have considered the association of drugs and proteins to capture the complex interactions in the biological network between drugs and proteins. To address this, we propose a novel framework leveraging knowledge graph-based meta-path learning (KG-DTA) to enhance DTA prediction. First, we construct a weighted heterogeneous network of drugs and proteins that incorporates three types of entities: drug-drug similarity, protein-protein similarity, and drug-protein affinity scores. Node features of both drugs and proteins in this network are enriched using embeddings from pretrained models. The meta-path learning technique extracts meaningful pathways within the network by focusing on paths from drug nodes to protein nodes, and a path length restricted to 2 or 3. This strategy yields 6 distinct pathways that capture intricate relationships between drugs and proteins. Finally, the XGBoost-based machine learning algorithm is used to predict the binding affinity score. Experimentally, we used two standard benchmark datasets: Davis and Kiba to evaluate KG-DTA. Experiments show that our method significantly surpasses the performance of state-of-the-art techniques, achieving mean square errors of 0.23 and 0.11 on the Davis and Kiba datasets, respectively. Additionally, our framework shows superior performance across four standard evaluation metrics, underscoring its effectiveness and reliability. These results suggest that knowledge graph-based meta-path learning is a promising approach to advance computational drug discovery, potentially accelerating drug development timelines.
Publication Source (Journal or Book title)
Proceedings of 4th International Conference on AI ml Systems Aimlsystems 2024
Recommended Citation
Ranjan, A., Bess, A., Sajol, M., Rajasekaran, M., Alvin, C., & Mukhopadhyay, S. (2025). KG-DTA: A knowledge graph-based meta-path learning framework to predict drug-target binding affinity. Proceedings of 4th International Conference on AI ml Systems Aimlsystems 2024 https://doi.org/10.1145/3703412.3703426